Possible to "miss" very short missions

[DaveyBiggers]

Most of our samples do something similar to this:

agent_host.startMission( my_mission, my_mission_record )
# Wait for start of mission (is_mission_running flag)
world_state = agent_host.getWorldState()
    while not world_state.is_mission_running:
        time.sleep(0.1)
        world_state = agent_host.getWorldState()

It's possible to create missions that can finish within 0.1 seconds - for example, in a setting where an agent must get to a goal square, but the start and end squares are randomised, it is possible for the agent's initial state to satisfy the end conditions.

In such cases, the above code can end up waiting forever, as the is_mission_running flag can be set and then unset again within the time.sleep() call.

One possible fix is to reduce the sleep time; alternatively, agent_host could have an is_mission_over flag too, instead of relying on is_mission_running.

[DaveyBiggers]

Note that this is only a problem once #23 is fixed, since without that fix very short missions break the mod long before they have a chance to break the agent.

[timhutton]

Yes. Maybe WorldState could replace is_missing_running with two bools, mission_started and mission_ended.

Then in user code this loop would be:

agent_host.startMission( my_mission, my_mission_record )
# Wait for start of mission (mission_started flag)
world_state = agent_host.getWorldState()
    while not world_state.mission_started:
        time.sleep(0.1)
        world_state = agent_host.getWorldState()

and the main loop would be:

while not world_state.mission_ended:
    ...

[DaveyBiggers]

Am making it a P4 since it will only be a problem for some mission designs, and there's an easy workaround in those cases (reducing the time.sleep value).

[timhutton]

Make startMission block until mission started? New API startMissionAndWait?

Samples should use peekWorldState too.